Mailing List Archives

Authenticated access

 		[Computer Systems Lab]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] SPOOL file clash with multiple submitters

After what seemed a huge amount of work I finally got this going. I pruned

out all of the stuff that I didn't need from the main Condor config file

then factored out all the common stuff and placed this into condor_config.local.

This way any "global" changes require only one file to be edited.

 

That left me with one condor_config for each scheduler which I managed to get

down to just e.g.:

 

LOCAL_CONFIG_FILE       = /opt1/submit/etc/condor_config.local

 

MASTER_LOG      = $(LOG)/MasterLog.Q1

SCHEDD_LOG      = $(LOG)/SchedLog.Q1

SHADOW_LOG      = $(LOG)/ShadowLog.Q1

 

SPOOL           = $(LOCAL_DIR)/spool/Q1

 

MASTER_ADDRESS_FILE = $(LOG)/.master_address.Q1

 

LOCK = $(LOG)/Q1

 

which has a nice simplicity to it. The lock file bit took a bit of working out.

 

If someone could get the "new" way of running multiple schedds working, I think this

would save of lot of effort - hint :-)

 

many thanks,

 

-ian.

 

 

 

From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Smith, Ian
Sent: 31 January 2012 10:29
To: 'Condor-Users Mail List'
Subject: Re: [Condor-users] SPOOL file clash with multiple submitters

 

Thanks Ian I will give this a go - I was thinking along these lines. As I see it, it

looks like the per-schedd environment doesn't get propagated properly to the shadows

so that they all pick up same the common S(SPOOL) directory. I should I have pointed

out that I'm using Condor 7.6.1 on Scientific Linux and this has a different to

spool area layout to 7.4.x (extra two layers presumably designed to limit the

number of files in each directory). I wonder if a bug crept in with the change or

if the problem was always there. Unfortunately I don't really have time to test

this with users chomping at the bit for the new server (always the way !).

 

regards,

 

-ian.

 

From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Ian Chesal
Sent: 30 January 2012 18:08
To: Condor-Users Mail List
Subject: Re: [Condor-users] SPOOL file clash with multiple submitters

 

On Friday, 27 January, 2012 at 9:57 AM, Smith, Ian wrote:

Thanks for this. As you say the nub of the matter seems to be how the different daemons

interpret the value of $(SPOOL) - particularly the scheduler, the negotiator and the shadows.

Can anyone from the Condor team shed any light on this - I can't find much info in the

manual.

You could try the "old" way of running multiple-schedds. Before the named instance stuff existed you had to give each schedd a unique condor_config file to read it's configuration from. It was convoluted, but it worked.

 

I'm pulling this from memory, so test this out before putting it on a production system!

 

Let's say you want to have two schedds on a machine. And for ease we'll say that you're running linux with the binaries in /opt/condor and the local directory in /opt/condor.local. You've got the following layout in /opt/condor.local:

 

/opt/condor.local/

         condor_config

         condor_config.local

         condor_config_schedd1

         condor_config_schedd2

         /1

                     /log

                     /spool

                     /execute

         /2

                     /log

                     /spool

                     /execute

         /log

         /spool

 

The condor_config file is your usual condor_config file that you'd use for your pool. You'll keep LOG set to $(LOCAL_DIR)/log in this file. That's where the MasterLog should go. The condor_config.local file defines only that there are two schedds on the machine:

 

         SCHEDD1 = $(SBIN)/condor_schedd.exe

         SCHEDD1_ENVIRONMENT = CONDOR_CONFIG=$(LOCAL_DIR)/condor_config_schedd1

         DAEMON_LIST = $(DAEMON_LIST), SCHEDD1

         DC_DAEMON_LIST = $(DC_DAEMON_LIST), SCHEDD1

 

         SCHEDD2 = $(SBIN)/condor_schedd.exe

         SCHEDD2_ENVIRONMENT = CONDOR_CONFIG=$(LOCAL_DIR)/condor_config_schedd2

         DAEMON_LIST = $(DAEMON_LIST), SCHEDD2

         DC_DAEMON_LIST = $(DC_DAEMON_LIST), SCHEDD2

 

The condor_config_schedd(1|2) files are complete copies of your condor_config file *plus* information specific to each of the schedd instances. So in condor_config_schedd1:

 

         SCHEDD_NUMBER = 1

         SCHEDD_NAME = $(SCHEDD_NUMBER)@$(FULL_HOSTNAME)

         SCHEDD_PREFIX = Q$(SCHEDD_NUMBER)

         LOG = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/log

         SPOOL = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/spool

         EXECUTE = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/execute

 

And in condor_config_schedd2:

 

         SCHEDD_NUMBER = 2

         SCHEDD_NAME = $(SCHEDD_NUMBER)@$(FULL_HOSTNAME)

         SCHEDD_PREFIX = Q$(SCHEDD_NUMBER)

         LOG = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/log

         SPOOL = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/spool

         EXECUTE = $(LOCAL_DIR)/$(SCHEDD_NUMBER)/execute

 

Obviously it's not nearly as elegant as the named config stuff, but that *should* work and might help with the propagation of the SPOOL setting down to the shadows for each named schedd on the system. 

 

You may not actually need complete copies of the condor_config file for the two schedd instance files (condor_config_schedd(1|2)) but I've never spent the time to whittle the schedd-specific files down to the bare minimum to see what works.

 

Regards,

- Ian

 

---

Ian Chesal

 

Cycle Computing, LLC

Leader in Open Compute Solutions for Clouds, Servers, and Desktops

Enterprise Condor Support and Management Tools